Understand the math behind every research paper.
Most AI tools summarize. Deconstructed renders every equation, explains it at your level, and computationally verifies it — so you actually understand the paper.
Equation-level explanations · verified by Wolfram · no signup to try
3.2 Attention
An attention function can be described as mapping a query and a set of key–value pairs to an output, where the query, keys, values, and output are all vectors. The output is computed as a weighted sum of the values.
Scaled Dot-Product Attention. The input consists of queries and keys of dimension d_k, and values of dimension d_v. We compute the dot products of the query with all keys, divide each by √d_k, and apply a softmax function to obtain the weights on the values.
The two most commonly used attention functions are additive attention and dot-product attention. Dot-product attention is much faster and more space-efficient in practice, since it can be implemented using highly optimized matrix multiplication code.
While for small values of d_k the two mechanisms perform similarly, additive attention outperforms dot-product attention without scaling for larger values…
Attention Is All You Need
Vaswani, Shazeer, Parmar, Uszkoreit, Jones · 21 sections
Recurrent neural networks have been firmly established as state-of-the-art approaches in sequence modeling and transduction problems.
The goal of reducing sequential computation forms the foundation of the Extended Neural GPU, ByteNet, and ConvS2S.
Most competitive neural sequence transduction models have an encoder–decoder structure.
An attention function maps a query and a set of key–value pairs to an output, computed as a weighted sum of the values.
This decides how much each word should pay attention to every other word. It scores how relevant the others are, turns those scores into percentages, then blends the other words together using those percentages.
The problem
You open a paper, get lost in dense equations, jump between tabs, ask scattered questions, and still don't fully understand what the math is doing.
You hit a wall of notation
You open a paper, equations pile up, and you lose the thread two pages in.
Tab-switching chaos
PDF in one tab, ChatGPT in another, notes in a third. Context scattered everywhere.
Summaries miss the math
Generic AI tools flatten the structure and skip the equations entirely.
You get the gist, not the understanding
You know what the paper claims, but not how the derivation actually works.
Why Deconstructed
Equation-aware explanations
Extract and explain the math — not just the paragraphs around it.
Built for technical papers
Understands sections, equations, and document structure without losing context.
Grounded in the source
Read the original paper alongside structured explanations. No more tab-switching.
From understanding to output
Turn one paper into summaries, flashcards, slides, and follow-up Q&A from the same workspace.
How it works
Upload a paper
Drop in a PDF or paste a URL. Deconstructed parses the structure, sections, and equations.
Explore the structure
Browse parsed sections. Every equation is extracted as LaTeX and rendered with KaTeX.
Dive into the hard parts
Click any equation for a Wolfram Alpha-powered deep dive — derivations, plots, and analysis.
Get explanations at your level
Basic intuition, advanced detail, or PhD-level rigor. Tailored to your background.
Turn insight into output
Generate summaries, slides, or flashcards from the same paper.
Upload a paper
Drop in a PDF or paste a URL. Deconstructed parses the structure, sections, and equations.
Explore the structure
Browse parsed sections. Every equation is extracted as LaTeX and rendered with KaTeX.
Dive into the hard parts
Click any equation for a Wolfram Alpha-powered deep dive — derivations, plots, and analysis.
Get explanations at your level
Basic intuition, advanced detail, or PhD-level rigor. Tailored to your background.
Turn insight into output
Generate summaries, slides, or flashcards from the same paper.
Try it now
Everything you get
Deep-Dive Math
Click any equation for Wolfram Alpha derivations, plots, and step-by-step breakdowns.
AI-Powered Parsing
Upload a PDF or paste a URL. Every section and formula is extracted automatically.
Interactive Explanations
Get conversational breakdowns of any section, with follow-up questions.
Difficulty Levels
Basic, Advanced, or PhD — explanations that match your background.
Study Artifacts
Generate summaries, slides, and flashcards from any paper.
Persona Customization
Set your field and goals for tailored explanations.
Cloud Sync & Sharing
Save papers across devices and share via link.
Saved Conversations
Every explanation is saved. Pick up where you left off.
Model Choice
Choose between Claude and GPT-5.4 for explanations.
Most paper tools stop at
“here's the gist.”
Deconstructed is for the moment after that — when you need to understand the notation, unpack the derivation, prepare for a presentation, or implement the idea yourself.
Summary tools
High-level overview
Equations skipped or paraphrased
Structure flattened
One output, one shot
Deconstructed
Equation-by-equation breakdown
LaTeX extracted and rendered
Original structure preserved
Explanations, deep dives, artifacts
Who it's for
Graduate students
Get through assigned papers faster and actually understand the equations before class, lab meeting, or your next presentation.
Researchers
Move into adjacent fields, digest unfamiliar methods, and break down dense math without losing the paper's structure.
Engineers & builders
Go from paper to implementation with a clearer view of what the model or method is doing.
Quantitative professionals
Read technical work in ML, statistics, economics, finance, and related fields with tools designed for formal reasoning.
For papers you can't just skim
Built for the fields where reading means doing math.
Pricing
Free
3 credits
on signup
- Haiku model
- Basic explanations
- Community support
Starter
5 credits
$1.40/paper
$7
- Sonnet, GPT-5.4, Gemini Pro
- Deep dives + Wolfram Alpha
- Summaries, slides & flashcards
Pro
15 credits
$1.27/paper
$19
- All models incl. Claude Opus
- Deep dives + Wolfram Alpha
- Summaries, slides & flashcards
Power
40 credits
$0.98/paper
$39
- Highest-volume paper workflows
- Deep dives + Wolfram Alpha
- Summaries, slides & flashcards
Have questions? Contact us
Stop wrestling with papers you don't really understand.
Deconstructed helps you work through the math, structure, and ideas — all from one workspace.
No credit card required. Free to get started.